The present disclosure relates to memory coherency, and more specifically, to improved machine performance with accelerator memory coherency.
Memory coherence provides for uniformity of shared data resources which may be stored in multiple locations, (e.g., local caches). If a data resource is stored in multiple locations, it is possible it may be modified in any of those locations. Memory coherency provides for ensuring modifications to shared data resources are centralized to a main memory.